How to Make a Training Video with Synthesia

Most “AI training video” posts skip the two facts that actually decide the project. First: a lesson is not a Friday update and it is not a policy memo. It is 75–120 seconds that names the one skill, walks the one form or screen, and closes on what people should do after they watch — open that form, submit the request, tick the checklist. Second: the tool this site sends for that talking-head file is Synthesia (synthesia.io). This guide is the working path when you will not film the trainer every time a field or an owner moves: SOP, host, one picture, generate, shelf.

When a how-we-do-this needs an avatar — and when it doesn't

Use an avatar when the LMS tile needs a face walking one procedure and you cannot or will not sit a named trainer in a room every time finance renames a field or ops moves an owner. Edit the line, swap the one slide or screen, generate again. That is the economic case.

Skip the avatar if L&D only needs a written SOP. Skip it if the skill has to be a live shadow — the lesson closes with “open this form” or “tick this checklist,” it does not replace a workshop. Skip it if the file is a Friday comms blast — the internal-update how-to is the other desk. Skip it if the file is a handbook amendment — the policy-update how-to reads a signed clause. Skip it if the file is a changelog — the release-notes how-to names what shipped. Skip it if you are buying a multi-lesson library with SCORM and quizzes — the AI-avatar training how-to is that desk. Skip it if the file is a public help-article path — the support how-to walks clicks for a customer, not an internal SOP.

The seven-step Synthesia training lesson

1. 01

   Write a 75–120 second lesson script from the signed SOP

   A training video is the procedure, spoken — and it ends on what the viewer does after the last line. Open on the one skill, name who the lesson is for, walk the one form or screen they will use, then close on the action: open that form, submit the request, tick the checklist in the LMS. Copy the words ops already signed. Time it out loud. Most single-procedure lessons sit between 75 and 120 seconds. If you wanted a Friday all-hands recap, that is the internal-update how-to. If you wanted a handbook amendment, that is the policy-update how-to. If you wanted a changelog read, that is the release-notes how-to. If you wanted a whole library with SCORM and quizzes, that is the AI-avatar training desk.

2. 02

   Confirm a talking-head lesson is the right format

   Use a presenter when the LMS tile needs a face walking one how-we-do-this and you will not film the trainer, the ops lead, or a camera every time the form or the owner moves. Skip the avatar for a live workshop, a manager shadow, or a wiki that does not need a voice. Skip it if the file is really a Friday comms blast, a signed handbook clause, a shipping-notes cut, or a public help-article click path. The avatar is a lesson host reading the current SOP, not a substitute for the live lab and not a substitute for a named trainer when the procedure is still in review.

3. 03

   Open Synthesia (synthesia.io) — check the domain

   Synthesia is the presenter desk this page is written for. On the home screen, start a blank video — Synthesia’s assistant help currently labels that Create video; some workspaces still show New video. Same start. Synthesys (synthesys.io) is a different company. HeyGen and a Loom-style recorder show up in the same “AI training video” search; they are editorial mentions here, not hops. Mixing the names is the expensive mistake. This page sends you to Synthesia only. The LMS is a shelf, not a hop.

4. 04

   Pick one stock avatar as the lesson host

   Select the Avatar tab in the top menu and choose one stock presenter. Keep that face for this SOP and for later recuts when a field or an owner moves. A lesson read is second-person and specific: you, this form, this next step. Head-and-shoulders. Do not commission a trainer likeness until L&D and Legal have watched a stock cut of the real procedure. Do not clone a trainer voice unless you have written consent. One host. One crop.

5. 05

   Add one slide or one screen pass — not a course tour

   The picture job is a single visual the avatar can point at: the form, the three fields, the one thing they will do after the clip. If the SOP is a slide you own, select Media at the top of the canvas and choose Upload media — Synthesia’s help lists images as .jpeg, .png, and .svg. Export that one lesson slide as an image and upload it. Import PowerPoint exists for a full .pptx; for one procedure slide, the image upload is enough. If the lesson only makes sense on a live internal screen, select Record at the top of the canvas, then Record Screen (Full Screen, Window, or Browser Tab), or Upload screen recording if you already have an MP4 — crop anything that should not leave the team. Place the visual beside or under a head-and-shoulders crop. We do not invent a “training” or “lesson” template. If your workspace has a saved layout, use it. If it does not, Media or Record plus the Avatar is the whole move.

6. 06

   Paste the script, then Generate

   Select a scene in the left scene menu, open the script box, and paste the spoken lesson. Split the walkthrough and the “what you do after this” beat across scenes if you want a pause on the form. Use Generate at the top right. If the file will play muted in an LMS tile, turn Burned-in captions on in that panel before you generate. Preview a scene first if you only need to check framing — Synthesia’s preview help says a preview does not use credits; Generate consumes minutes. Watch the first second, the procedure name, the one visual, and the close that names the next action. Mouth-sync is consistently strong in English and major European languages. Hand gestures loop. These are narrators. Edit the line and generate again instead of re-booking the trainer.

7. 07

   Download the MP4 or copy the share link, then drop it in the LMS above the action

   A clean download starts on a paid plan — Synthesia’s download help says Basic and trial users cannot download. After generate, use the three-dot menu and choose Download, then the MP4 (Full HD 1080p). Or open Share, create a link, and copy the link or the embed code; Synthesia’s share help says an embed updates when you edit. Upload the MP4 into the LMS module L&D already assigns, or paste the share link / embed if that LMS accepts it. Put the same next-step line in the module intro that you put on the slide: open the form, submit the request, tick the checklist. Keep the written SOP under the player. Name the export so the next recut is obvious: lesson-file-escalation_aug_v1. When a field or an owner moves, change that line, generate, bump the version, replace the file or the link.

Lesson-host avatar vs filming the trainer vs a written SOP

“AI training video” is a search, not a product. The decision is the format. Plan names and download gates on the Synthesia column match the live synthesia.io/pricing page as of August 2026. Confirm the current grid before you tell L&D the export is included.

A 75–120s lesson script is the SOP, spoken — then the next action

Do not write a course outline and hope someone cuts it down. Open the signed SOP, the form the team actually uses, and the LMS assignment L&D will attach: the one skill, who it is for, the three fields or clicks that matter, the one thing they will do after the last line. Those sentences are the script. If a field is still in review, that clip waits. An avatar walking last quarter’s form is worse than a wiki, because it sounds finished.

Four beats, in this order. Name the lesson in the first sentence. Say who it is for in one line. Walk the one form or screen against the one picture. Close on one action they take after they watch: open that form, submit the request, tick the checklist in that same LMS module. Second person. No joke about “learning being fun now.” Time it out loud: 75 seconds is a tight procedure; 120 seconds is a procedure plus one “what this is not” beat. Longer than that and you are making a welcome or a course.

A customer-escalation lesson that fits the window sounds like this — about 90 seconds when you say it, not when you skim it:

This is the lesson on how we file a customer escalation. After this clip you open the escalation form and submit it before you Slack the account team. Use this when a customer is blocked, an SLA is already late, or a deal owner asked for a named owner. Do not use this for a how-to ticket — that stays in the help article. Three fields: severity, the account, and what you already tried. The screen in this clip is the whole form. Open the form linked under this video and file it. If you already pinged Slack, go back and file the form — the thread is not the record.

An expense-submit or a production-access request uses the same skeleton. Name the skill, name who it is for, point at one slide or one screen, close on the form they open after the clip. Do not stack a culture speech, a six-slide course tour, and a quiz in a single file. One lesson, one picture, one next step.

Do not write a Friday Slack recap. Do not write a handbook clause with an acknowledgment. Do not write a shipping list. Do not write a trial CTA card. Do not ask the avatar to “sound like the trainer.” These presenters are narrators: blink cadence, micro head-tilt, a looping hand. On a head-and-shoulders framing that is enough. Full breakdown of what the avatars actually do is in our Synthesia review.

Pick one lesson host. Keep that face.

Select the Avatar tab in the top menu. Choose one stock presenter and keep that face when you recut the series so the LMS does not look like a new hire every SOP. A lesson host says you and this form. It is not a comms host reading Friday’s brief, not a policy host walking a handbook, and not a release host reading a changelog.

A custom trainer likeness is a later decision, not the first invoice. Prove the script on a stock presenter. Personal avatars sit on paid plans — confirm the current count on synthesia.io/pricing. Synthesia’s pricing grid lists a Studio Express-1 avatar as a $1,000/year add-on that can take up to ten days. Clone a voice only with written consent. The consent rule is the same one in How to clone your voice for YouTube.

One slide or one screen pass — Media, Record, or an image export

People watch these muted, on a laptop, in an LMS tile. The voice is not enough. Pick one visual the avatar can point at for the whole cut: a still of the form, or one screen pass of the same form. Not both. Not a tour.

If the procedure is a slide you own, select Media at the top of the canvas and choose Upload media. Synthesia’s media help lists images as .jpeg, .png, and .svg, and videos as .mp4 or .webm. Export one lesson slide from the SOP as an image and upload that. Import PowerPoint is a home-screen path for a full .pptx — useful when you already have the deck, easy to overshoot when you only need one frame. We do not invent a “training lesson” control. If your workspace has a saved layout, use it. If it does not, this upload is the whole move.

If the lesson only makes sense on a live internal screen — the escalation form, the expense tool, the access request — select Record at the top of the canvas. Synthesia’s screen-recorder help lists Record Screen (Full Screen, Window, or Browser Tab) and Upload screen recording for an existing MP4. Crop the chrome you do not want in a company-wide module. You are showing the one form, not walking the whole admin console. A how-we-do-this lesson is usually this pass, not a six-slide deck.

Optional on-screen label: select Text at the top of the canvas and pick Title 1 or Title 2 for the lesson name, Subtitle or Body for the next action. Type the same words the avatar will say. Drag a corner to size it; keep it readable on a narrow LMS tile. Do not stack the whole SOP on the card.

Captions: Synthesia’s help lists closed captions, dynamic captions (Captions in the top menu), and a Burned-in captions toggle in the Generate panel. For a muted LMS tile, turn burned-in captions on before you generate. After generate you can also download .srt or .vtt from the video menu if the LMS wants a separate track.

Generate, then export into the LMS above the action

Generate sits at the top right. Title the file the way the LMS module is titled. Watch the first second, the one visual, and the close that names what they do after the clip. Proper nouns, field names, and owner names are where these reads go wrong — fix the script, generate again. That loop is the product. A trainer re-book is what you are not buying.

A logo-off download starts on Starter. After the render, open the three-dot menu and choose Download, then the MP4 (Synthesia’s help lists Full HD 1080p). Or open Share, create a link, and copy the link or the embed code. The share link is the better shelf when you will recut the same series; the MP4 is the better shelf when the LMS wants a file. Interactive buttons, branching, and quizzes cannot go out as an MP4 — leave those off a lesson you need to download. If a last scene still points at a “next scene” instead of ending the video, Synthesia’s download help says the MP4 can stay locked; set that last scene to end, generate again.

Upload the MP4 into the LMS track L&D already assigns — Workday Learning, Lattice, or a generic LMS — or paste the share link or embed if that editor accepts it. Put the same next-step line in the module intro that you put on the slide. Keep the written SOP under the player. A viewer who cannot play video still needs the sentences. The LMS is an ordinary destination on this page, not a hop — use whatever control that editor already has for an upload or a link. We do not invent a “Synthesia app” inside those products. SCORM export sits on Enterprise; buy that tier when L&D asks for a package, not for a single English lesson you can upload as an MP4. Quizzes and branching live on Creator and above — they are a later desk, and they are the library training how-to, not this brief.

Name the object so the next person can find it: lesson-file-escalation_aug_v1, lesson-submit-expense_sep_v1. When a field or an owner changes, edit that line, generate, bump the version, replace the file or the link. Leave the clips that are still true.

Which Synthesia plan ships a training lesson in 2026

The live pricing page (August 2026) lists Basic at $0 with 9 avatars and 10 minutes a month. Downloads and logo removal start on Starter at $29/month, or $18/month billed yearly. Creator is $89/month, or $64/month yearly, with more minutes and personal avatars. Languages and voices are published as 160+ on every tier. Confirm the current grid on synthesia.io/pricing before you tell L&D the download is included — we do not invent a checkout total or a commission rate.

For a first lesson, start on Basic with a real SOP paragraph, not a placeholder greeting. Move to a paid plan when you need a logo-off file or a link you will actually assign. Buy Creator for a personal lesson-host avatar, more minutes, or interactive video — not for “more cinema.” Buy Enterprise when brand asks for a kit, IT asks for SSO, or you are ready to localize a locked English master. Confirm internal-use terms on the plan you will export from. Ten minutes a month is a handful of 90-second lessons, not a year of SOP recuts.

Localization and other desks are next steps, not this brief

Ship the home-market lesson first. A second language is a translation of a signed SOP, not a reason to start in Enterprise. When you are ready, lock this cut, then follow how to localize training videos — same loop, different packet. A Friday all-hands recap is the internal-update how-to. A handbook amendment is the policy-update how-to. A shipping-notes cut is the release-notes how-to. A multi-lesson library with SCORM and quizzes is the AI-avatar training desk. A voice-only lesson with no face is training voiceovers with Murf — editorial, not a hop on this page.

Learners still deserve a plain-language note when the presenter is synthetic. Say it in the LMS intro, above or under the player: this is a lesson-host avatar walking the current signed SOP. If you later publish the same file on a public platform, you also need that platform’s current disclosure for synthetic media. That is separate from the vendor license. You need both.

The comparison of presenter platforms — including where Synthesys sits as a different company — lives on Synthesia vs Synthesys and Best AI Avatar Video Tools 2026. This page does not hop there. The only Try button here is Synthesia.

HeyGen, a Loom, and a filmed trainer — editorial only

Those names sit in the same “AI training video” search. HeyGen is the usual Synthesia alternative in reviewer write-ups: faster custom-avatar turnaround, punchier social presenters, less of the LMS/SSO shelf. A Loom or a native screen recorder is the right tool when you do not need a face. A filmed trainer is the right tool when the face has to be that person and the file will not be recut. None of them have a live hop on this page, so there is no Try button. If you need the single-lesson cut we can route, it is Synthesia.